An NLP project using Word2Vec embeddings to analyze and recommend Twitch emotes in chat messages. Files are named and commented in tutorial fashion. This repository will walk you through step-by-step how to scrape your own Twitch chat data, analyze it, train a Word2Vec model with it, and use it to make predictions and recommendations. The purpose of the recommender is to help a chatter understand emotes in a new channel based on the emotes already available to them (global emotes, subscribed emotes, 3rd party apps like BTTV and FFZ). The data used for this project has been uploaded to the DOI listed below, including emote registries, chat logs, trained model embeddings, and labeled metadata compatible with TensorFlow's Embedding Projector.
Data available here: 10.5281/zenodo.8012284